基于图像特征的隧道运动火源视频火焰探测技术分析-analysis of video flame detection technology for fire source in tunnel movement based on image features.docxVIP

摘要隧道交通的不断建设，对于缩短里程、提高行车效率具有突出的贡献。然而， 目前隧道交通的行车速度、行车密度不断加大，隧道火灾事故频发，特别是运动 车辆引发的火灾造成的后果更为严重。隧道是一个半封闭的受限空间，且受其中 粉尘多、湿度较大和纵向风的影响，传统的火灾探测技术显示出力所不能及的一 面。本文结合目前发展较好的火焰视频探测技术，针对隧道中火源在运动条件下 的火焰视频进行相关的探测技术研究。本文在实体实验的基础上，对运动火源的火焰视频数据进行采集，同时为了 与火焰视频数据形成对比，对静止的钨丝灯、运动的车灯等视频数据进行了采集。 结合火焰的颜色和亮度信息，建立了针对隧道这一特殊环境背景的更新模型。这 一模型对于去除隧道环境中少量太阳光反射和壁面对火焰亮光反射的影响，其效 果较好。但是采用这一方法，静止的钨丝灯和运动的灯光被作为火焰候选区域显 示出来。因此，本文在提取候选区域的基础上，结合火焰在燃烧过程中不断抖动，其 边缘的尖角数量不断变化，区别于车灯等其边缘较为稳定的情况。本文采 Harris 算法对图像中的尖角进行提取，并通过对帧间尖角数量的变化阈值的研究，对火 焰图像和非火焰图像进行识别。由于火焰闪烁频率和火源的尺寸等相关，所以直接依据火焰的闪烁频率难以 实现对火焰区域和非火焰区域的识别。本文建立了基于火焰闪烁的火焰识别的补 充判据。通过对候选区域面积的研究，发现其火焰的面积随时间不断振荡，而静 止的钨丝灯面积基本保持不变，运动的车灯的面积随时间有小幅度的增长。本文 对视频图像中，连续的三帧图像中相邻两帧图像面积差的乘积的阈值的分析研 究，最后通过阈值的设定，识别火焰图像和车灯图像。关键词：隧道火灾 运动火源 运动目标检测 颜色特征 火焰尖角 频闪特性 傅里 叶变换ABSTRACTThe continuous construction of tunnel traffic is outstanding contribution to shorten the distance and improve traffic efficiency. While, with increasing of tunnel traffic speed and traffic density, tunnel fire accidents are more frequent, especially accidentscaused by vehicles. As a semi-enclosed and limited space, it is dusty，humidity andlarge vertical wind in tunnel. The traditional detection technology is not applicable in tunnel. In this paper, the video flame detection technology for moving tunnel fire is investigated.Based on the experiment, flame video data of moving tunnel fire was collected. Meanwhile, video data of stationary tungsten and lamp was obtained, in order to compare the flame video data. According to flame color and brightness information, a background update model is established for tunnel. the model is effective to remove a small amount of reflected sunlight and flame. However, stationary tungsten and lamp are identified as candidate region of the flame.Therefore, based on the candidate region, the flame continuous shake and the changing number of the cusp is investigated, which is different from the stable of the edge lights. The Harris algorithm is adopted to extract the flame cusp. Then, changes in